I'm betting there's a feature that corresponds to orderings: numbers, letters, taxonomies, etc. Are there things LLMs internally order that we'd be surprised about? Can we use a residual stream feature to look for it?
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{holtzman-implicit-orderings-2026,
author = {Holtzman, Ari},
title = {Implicit Orderings},
year = {2026},
url = {https://hypogenic.ai/ideahub/idea/HzutqK2wTeelMMzNcLc7}
}Please sign in to comment on this idea.
No comments yet. Be the first to share your thoughts!